The Busara Center: A Laboratory Environment for Developing Countries

Johannes Haushofer

Harvard, MIT

Marie Collins, Giovanna de Giusti, Joseph Muiruri Njoroge, Amos Odero, and Cynthia Onyango

Innovations for Poverty Action, Kenya

October 1, 2012

Abstract
While Randomized Controlled Trials and lab-in-the-eld experiments have become more common in development economics in recent years, the rigor and control of laboratory-based experiments has so far been dicult to access in developing countries. Here we summarize the creation of the Busara Center for Behavioral Economics, a state-of-the-art laboratory for behavioral and experimental economics in Nairobi, Kenya. We provide data on physical and technical setup and infrastructure, protocols for study administration, respondent and data ow, and subject recruitment, payment, and subject pool composition, and outline how researchers can use the lab. JEL Codes : C93, D03, D87, O12 Keywords: development, laboratory experiments, experimental economics, behavioral economics, psychology
Our gratitude for making Busara possible goes to the sta of Innovations for Poverty Action, in particular, Karen Levy, Colin Christensen, Daniele Ressler, Delia Welsh, Niall Keleher, Justin Oliver, Nicole Grant, Joann Phelan, and Annie Duo; the sta at the University of Zrich, in particular, Stefan Schmid, Michael Hohl, and Adrian Etter. We thank Ernst Fehr, David Laibson, Iris Bohnet, Mark Edington, Amal Devani, and Jasper Grosskurth for valuable feedback and advice; Anne Kioi, Japeth Juma, Febrone Atieno, Anne Wynne, Carol Aluchio, Nonnie Kamau, Lillian Achibeala, and Peter Gachanja for research support; and Anne Healy for comments on the manuscript. joha@mit.edu

Electronic copy available at: http://ssrn.com/abstract=2155217

Introduction

Since their beginnings in the 1960s, laboratory experiments have become an increasingly popular tool for studying economic behavior in controlled settings. In his early double-acution experiments, Vernon Smith (e.g. Smith 1962; Smith 1982) showed that core features of microeconomic theory can be tested and proven accurate in laboratory experiments, and thus provided an early proof-of-concept for the usefulness of laboratory experiments in economics. Since these early successes, experimental economics has become a large and productive subdiscipline whose reach is still expanding; for instance, recent research has begun to model macroeconomic phenomena in laboratory settings (e.g. Lei and Noussair 2002). The rise of laboratory experiments in economics also prepared the ground for the rise of behavioral economics, notably through the work of Amos Tversky and Daniel Kahneman (Kahneman and Tversky 1979). The goal of this reseach program was to injeect psychological realism into models of economic behavior, and it relied heavily on laboratory experiments to do so. In addition, and partly owing to inspiration from behavioral economics, recent years have seen a surge of interest among psychologists, cognitive scientists, and neuroscientists in the cognitive and neural processes underlying reward, valuation, and decision-making (e.g. Schultz 2006; Glimcher 2003; Glimcher 2009); these studies, too, are mostly conducted in lab settings. Together, these developments within experimental, behavioral, and neuroeconomics have ushered in a new era of psychologically inspired research on economic preferences and decision-making in the laboratory, and today these elds are among the most prolic and rapidly evolving subdisciplines of economics. Surprisingly, however, laboratory studies in behavioral science have so far largely been restricted to developed countries, frequently using university students as subjects. In psychology, for instance, a 2008 survey by Arnett found that 96% of subjects in studies published in top journals were from "WEIRD" (Western, educated, industrialized, rich, and democratic) backgrounds. This fact represents a signicant challenge to the external validity of the

ndings obtained in these contexts. Indeed, those experiments that did study non-Western subjects often found considerable variability across cultures (Henrich et al. 2001; Cardenas and Carpenter 2008). These results, in turn, led to calls for more research on the behavior and preferences of people from non-Western contexts (Henrich et al. 2010a; Henrich et al. 2010b). Despite these developments, to our knowledge the rigor and control of laboratory environments have not been readily accessible in developing countries to date. We report here on the creation and features of the Busara Center for Behavioral Economics in Nairobi, Kenya. The core of Busara is state-of-the-art laboratory with 20 networked PCs, combined with a

Electronic copy available at: http://ssrn.com/abstract=2155217

subject pool of around 2000 residents of the Nairobi slums (an additional subject pool from the University of Nairobi is being recruited at present). Use of touchscreen computers and special paradigms enables even illiterate respondents to participate. Busara can be used by researchers around the world; its goal is to provide economists, psychologists and cognitive scientists, political scientists, anthropologists, and other behavioral scientists with an infrastructure which facilitates the administration of high-quality laboratory experiments. The

goal of this paper is twofold: rst, it aims to provide enough detail to enable others to create similar labs in other countries; in the long run, we hope that Busara will become a node in a network of social science laboratories in developing countries. Second, the paper aims to serve as a reference and manual for researchers interested in running studies at Busara.

Advantages of laboratories

Before we describe the setup and features of Busara, we briey list the advantages a laboratory setting has over eld or lab-in-the-eld experiments. In eld experiments, data is typically collected when a eld ocer conducts one-on-one interviews with respondents, either with a paper questionnaire or with a laptop computer. Lab-in-the-eld experiments

(cf. Cardenas and Carpenter 2008) attempt to establish laboratory-like conditions in eld settings  e.g., participants might be gathered in the classroom of a local school to play a trust game using paper-and-pencil communication. In our view, dedicated laboratories have several advantages over such settings, which we briey describe in the following. First, lab settings allow researchers to maintain the anonymity of participants and their responses to a much greater degree than eld or lab-in-the-eld experiments: in the latter forms of experiment, participants often know the other participants and their choices in an experiment, either because choices have to be made publicly (especially when participants are illiterate), or because the setting is such that even private decisoins become public immediately due to the diculty of restricting communication between participants during the experiment. In contrast, lab settings allow participants to be seated in private cubicles and giving responses anonymously. A related advantage is that lab settings allow for sensitive information to be communicated condentially, with neither the researcher nor other participants becoming aware of a subject's response to any given item; in eld settings this is often not possible, mainly because responses are recorded by eld ocers. Second, eld or lab-in-the-eld are frequently conducted in small communities, where participation is public knowledge and many respondents in a particular session may know each other. While such familiarity may be useful or necessary for some experiments, it is more desirable to induce it through laboratory techniques in a sample of unacquainted re-

spondents (e.g.

provision of selected pieces of information about other participants, or a

minimal group paradigm) than to use the endogenously formed familiarity patterns already present in a sample of acquaintances and family members. Economics labs make this approach possible, not only because they make it easier to manipulate perceived familiarity among participants through experimental techniques, but also because they often have large subject pools of potential participants, drawn from geographically disparate areas. Selecting a random sample from such pools, especially when it happens privately such as through text messages, enables researchers to ensure that a minimum number of participants know each other before the study. Third, computerized lab experiments allow researchers to conduct interactive games with much greater ease and eciency than eld or lab-in-the-eld settings. Interactive economic exchange games, such as Trust, Dictator, Ultimatum, Public Goods Games, or Double Auctions, are backbone paradigms of experimental and behavioral economics, but they are cumbersome to implement as paper-and-pencil experiments in eld or lab-in-the-eld settings, whereas their administration is easy and fast in lab settings with networked computers. Forth, and relatedly, a computer lab allows for richer stimulus sets than are possible in paper-and-pencil experiments. For instance, researchers can easily employ pictorial, audio, and video stimuli, in e.g. priming studies. In paper-and-pencil experiments, this is either not possible or more dicult. A fth and related advantage concerns respondent comprehension. This is a key concern for economic experiments, especially in developing countries with low education levels, and computerized administration makes it easier for researchers to facilitate and test understanding with animated instruction videos, practice rounds, and test questions. Sixth, while paper-and-pencil experiments usually require at least basic literacy, computerized lab experiments in developing countries allow for the participation of both computerilliterate and entirely illiterate respondents. In particular, we describe below how we use

touchscreens to enable participants without familiarity with computers to participate, and how specially developed paradigms can make lab studies accessible to entirely illiterate participants. Seventh, computerized lab studies obviate the need for data entry. For paper-and-pencil studies, data entry can be a signicant cost` in terms of time and money, and an important source of errors; in computerized studies this factor is eliminated, increasing data quality and decreasing cost. Eigth, computerized lab experiments are ecient; data can be collected from many participants at once, with a small number of research assistants (RAs). This is generally not possible in survey experiments, which are often face-to-face.

Finally, Busara has important advantages over internet data collection through tools like Amazon's Mechanical Turk (MTurk). In recent years, such internet studies have rapidly

gained popularity, especially in the psychology community. Platform like MTurk allow researchers to put experimental paradigms online, and thus rapidly gather data from hundreds of subjects whose collection would otherwise take a much longer time in the lab. However, a critical limitation of studies on MTurk is that they require a) access to a computer, b) basic reading and, often, writing skills, c) a bank account or other formal nancial mechanism for receiving payment. The large majority of the world's population fails to meet any one of these criteria, let alone their combination. Thus, MTurk studies still exclude most of the

world's population, especially the poorest. In contrast, to participate in studies at Busara, participants do not need a computer, reading/writing skills, or a bank account; the only requirements are that participants be 18 years or older and have access to a cell phone and the MPesa mobile money system. Recent data collected in the Nairobi slums by Tavneet Suri shows that, despite their poverty, more than 90% of people there have both a cell phone and MPesa access. Thus, Busara is set up to include even the poorest participants, and thus provides a unique opportunity to study respondents at the very bottom of the pyramid, which are not accessible through internet tools such as MTurk.

Physical infrastructure

The Busara Center for Behavioral Economics is located in the Kilimani neighborhood of Nairobi, Kenya, within the premises of Innovations for Poverty Action Kenya (IPA-K). Kilmiani is a relatively safe and upscale neighborhood and is home to many NGOs. Nevertheless, it is also located one mile from Kibera, Nairobi's largest slum. The compound has a 24-hour security service, with two guards at the gate at all times, who have access to radios and panic buttons for calling backup in the case of security threats. Busara oers a state-of-the-art infrastructure for experiments in behavioral economics and other experimental social sciences. The facilities available include: a networked computer cluster; a waiting room; four individual survey cubicles for private interviews; laboratory space for saliva and blood sampling; and an oce with several desks for sta, visiting researchers, and students. 1. The networked computer cluster consists of 20 workplaces for participants that are equipped with 20 computers, and one additional control computer at the front of the room. All computers are Hewlett Packard TouchSmart 310 All-in-One PCs, equipped with keyboard, mouse, camera, microphone, and headphones (described in more detail below). A central feature of the computer setup is that all computers have touchscreen

monitors, and therefore can be used by illiterate subjects or subjects who are unfamiliar with operating a mouse or keyboard. The computers are arranged in four rows of 5; each individual desk is 70 cm wide and 60 cm deep, and is separated from the neighboring desks through dividing panels that provide complete privacy and anonymity while performing tasks. The panels are 60 cm high on three sides of the desk, and extend 30 cm beyond the edge of the desk in the horizontal direction, to ensure that participants cannot observe each other from the side. To turn the laboratory into a classroom, the dividing panels can be fully removed, without tools, by simply sliding them out of guide grooves and stacking them on the side of the room. The cubicles are numbered

busara01-busara20,

and each cubicle is tted with a sign

indicating its number. When participants arrive for a study, they are randomly assigned a particular cubicle and given a placecard that matches that of the cubicle. The

placecards are marked not only with the cubicle names, but also with 4 dierent colors (one for each row: Red, Green, Blue, Yellow), and 5 dierent symbols (circle, eye-shape, triangle, square, and star for the rst to fth PC in each row). Together, the colors and symbols of the placecards uniquely identify the cubicles, allowing even illiterate participants to nd their assigned seat. 2. A large oce provides 6 desks for sta as well as visiting investigators and students; in addition, it hosues a conference table which can be used for meetings, and an additional Hewlett Packard TouchScmart 310 All-in-One PC, with the same conguration as the computers in the main lab, that can be used for developing and testing programs and paradigms. 3. A 20-person waiting room accommodates respondents while they wait, and for briengs before and after studies. 4. An additional room provides 4 individual survey cubicles in which participants can be interviewed one-on-one, with complete privacy. The walls of the individual cubicles extend from oor to ceiling, creating completely closed and separate rooms. Each room has both natural and artical light and is air-conditioned. 5. Finally, Busara houses a small laboratory equipped for obtaining saliva and blood samples. A -20 deg C freezer is available on-site for specimen storage; continuous and uninterrupted power supply is ensured through a backup system for power outages up to 8 hours. In addition, a -80 deg C freezer owned by Busara is located at Kenyatta National Hospital, about 1 mile from Busara; this freezer is attached to Nairobi's

hospital power grid and is therefore not aected by power outages. A phlebotomist with a degree in laboratory sciences from the University of Nairobi is available to perform blood draws using the Vacutainer system. Supplies for saliva samples and blood draws are available locally, and a stock is kept in the Center. Analysis capabilities for saliva and blood samples (e.g. cortisol, CRP, cytokines) are available at Lancet Labs, a

state-of-the-art commercial laboratory and long-time Busara partner, located half a mile from the Center.

Technical infrastructure

4.1 Hardware
4.1.1 Lab computers
The 22 computers at Busara are Hewlett Packard TouchSmart 310-1126 All-in-One high performance PCs. Each computer has a 1TB harddrive, 6 GB RAM, an Intel Core 2 Duo processor at 2.90 GHz, a webcam, a CD/DVD and Blu-Ray player and burner, Bluetooth capability, several USB ports, a media card reader, a wireless network card and ethernet ports, high-quality speakers, a headset with microphone, and a 20-inch widescreen touchsensitive display. The PCs are all-in-one machines, i.e. all hardware is integrated into the screen, obviating the need for placing a tower case by respondents' feet. In addition, a 1.5 TB network attached storage system (NAS) is available; it is mounted on all computers and used for data storage and common access, as well as backup of all experimental and subject pool data. In particular, the NAS allows all experiment PCs to access the same les, e.g. images to be used in experiments, or batch les to start programs. Thus, if a batch le

involved in starting up the lab needs to be changed, it is sucient that this change be made once. The entire lab is fully networked using both wired and wireless connections, to create redundancy in case of the failure of one system. Network connectivity is provided by a 48port switch located in a server room. A total of 36 active

wired

connections are available, of

which 24 are allocated to the lab computers; 4 are available in the individual survey cubicles; and 8 are available in the researchers' oce. In addition,

wireless

network connectivity is

provided using DHCP throughout the entire lab, allowing for a large number of simultaneous connections. All PCs are assigned xed IP addresses in a dedicated IP address range on both the wired and wireless connection. The building is connected to the internet through a highspeed beroptic line; the arrival of beroptic internet in Kenya in 2009 has dramatically increased attainable connection speeds and bandwidths: download speeds at Busara are

typically 2 Mbps, and upload speeds 0.5 Mbps.

These speeds allow, for instance, for the

unproblematic use of YouTube videos as stimuli in studies. Electricity is supplied by Nairobi City Council and backed up through an uninterrupted power supply system, which can provide power to the lab for 8 hours during power outages. The TouchSmart PCs are distributed as follows: 20 PCs serve as client computers in the main computer lab (computer names: computer at the front of the lab

busara01-busara20) (busaracontrol)

; one PC serves as the control

; and another PC is located in the researchers'

oce and serves as a backup machine and testing computer for developing programs and paradigms

(busaraoce)

The cubicles are labeled with printed and laminated A5-sized

signs with the number of the computer, a color, and a symbol, which uniquely identify the computer as described above.

4.1.2 Enrollment and Identication Netbooks, Fingerprint Scanners, and Recruitment Modems
In addition to the lab infrastructure described above, we use three Hamster IV nerprint scanners from 360 Biometrics, in combination with three Asus EEE Netbooks, to record participants' ngerprints when they rst sign up for the subject pool, and to identify them when they come to the lab for sessions for which they were invited by SMS. Automatized sending of bulk SMS invites is performed with a laptop equipped with a Huawei USB modem containing a Safaricom SIM card.

4.2 Software
4.2.1 Lab computers
All lab computers at Busara run Windows 7 Home Edition and Norton Antivirus. All notications, automatic updates, and screensavers are disabled to facilitate optimal performance of the experimental software and avoid potential distraction or interruptions of study sessions with participants. Any necessary operating system updates are performed manually once per week. The NAS is mounted on all computers. The NAS is also set up for VPN access, allowing researchers to access it remotely with a password. The main software used for interactive economic exchange games is zTree (Fischbacher ). This software, also developed at the University of Zrich, is specially designed for the

creation and administration of computerized economic experiments. The main instance of the program is run by the experimenter on the

busaracontrol

machine, and client programs

("zLeaves") are opened automatically upon startup of the client PCs. The connection between the main zTree instance on

busaracontrol
8

and the client computers is established via

the IP address of the server; thus, a subset or all of the client computers could in principle be located in a dierent lab (e.g. in Europe or the US). The computers are started and shut down by using LabControl, a software developed at the University of Zrich, which uses Wake-on-LAN to remotely boot machines. Upon

startup, all machines execute a batch le that opens a zLeaf and thus the connection to the host computer. The successful startup of both the computers and the zLeaves can be monitored on

busaracontrol.

Thus, it is possible to make the entire lab ready for running an

experiment with one click. It is possible to run more than one experiment during a study session by running more than one instance of zTree on

zTree instances using zTree's built-in

busaracontrol

busaracontrol channel

, and connecting subsets of zLeaves to dierent feature. The dierent instances of zTree on

are distinguished by using dierent channels, and the zLeaves and the zTree

that belong to one experiment must use the same channel. In the course of each experiment, the session data is stored in an Excel le, which can easily be used for data analysis. zTree is congured such that output les are writting to a password-protected directory on the NAS, and thus are accessible through VPN.

4.2.2 Subject Database, Enrollment, Identication, Study Signup, and Respondent Payment
On recruitment days in the slums and at the University of Nairobi, participants' demographic information is registered using the Blaise computer-assisted interviewing software, which runs on two Asus EEE Netbooks (Enrollment Netbooks). Blaise is interfaced with an MS Access Database on these netbooks; at the end of a recruitment day, the databases from the two Enrollment Netbooks are consolidated and stored on a desktop PC in the Busara oces, as well as backed up on the NAS. In addition, each newly registered participant provides a ngerprint during registration; this ngerprint is recorded with a Hamster IV scanner connected to the Enrollment Netbooks using the PersonID software by 360 Biometrics. Each respondent's ngerprint is linked to their demographic data in the MS Access database through a unique identier; this identier is assigned automatically to both the ngerprint and the demographic data, thus precluding possible errors from manual entry. When subjects report for a study session at the lab, their ngerprints are scanned at the gate of the Busara compound using an additional Hamster IV scanner connected to another Asus EEE Netbook (Identify Netbook). The unique identied for the subject is automatically populated into a eld in a Blaise program called Identify, which then retrieves that subject's name and age from the MS Access database. This information is then used to conrm that the subject appears on the list of invitees for the particular session. The Blaise

program then records the attendance of the subjet pool member at the particular session; this information is later added to the MS Access subject pool database, and can subsequently be used to a) track session attendance by respondent, study, and signup mechanism (phone vs. SMS), b) selectively exclude participants from participation in particular studies depending on their participation in earlier studies. E.g., researchers conducting a study on trust might want to exclude participants who have previously played a Trust Game at Busara; the tracking of attendance and participation through ngerprints ensures that the subject pool contains up-to-date information on the experimental history of all respondents in the subject pool. The list of invitees for a particular session is generated as follows. First, a Stata script draws a random list of respondents from the subject pool database, according to the demographic criteria specied for the particular study (e.g. age, education), and prior study participation (see above); second, the bulk SMS sofware FrontlineSMS, running on a laptop equipped with a Huawei USB modem and a Safaricom SIM card, automatically sends personalized SMS messages to all respondents on this invite list. Respondents reply to these SMS messages with the word "YES" to register for a particular session, or "NO" to decline the invitation. FrontlineSMS automatically decodes these responses and generates a "YES"-list for each individual lab session, which is then used at the gate for respondent identication. Finally, after a session is complete, subjects are paid through an MPesa transfer to their cell phones. This is achieved as follows: during the experiment, zTree (or another survey software) keeps track of subjects' earnings in the tasks performed by subjets. At the end of this experiment, a 200 KES (~USD 3) show-up fee is added to this amount, as well as a KES 50 (~USD 0.60) bonus if the participant arrived on time. A Stata script then

generates a payment list including names, phone numbers, and amounts to be transferred, for all participants in the session. This le is sent to the IPA-K nance team, who initiate the MPesa transfer to respondents' cell phones. The time between completion of an experimental session and the arrival of the payment in participants' MPesa accounts is 2 hours on average.

Subject pool recruitment

Busara maintains a subject pool with currently around 2000 members, who are recruited from the Nairobi slums and the University of Nairobi. Permission for recruitment was obtained from the District Commissioners (DCs) of Langata and Embakasi districts of Nairobi; in addition, Busara's activities are covered by Institutional Review Board (IRB) approvals at the Massachusetts Institute of Technology (MIT), Harvard University, Innovations for Poverty Action-Kenya (IPA-K), and the Kenya Medical Research Institute (KEMRI). In addition to

10

subject recruitment, these IRBs also cover many standard behavioral games. Researchers running studies at Busara need to obtain IRB approval from their home university. Before carrying out recruitment for the respondent database in a particular area, Busara representatives met local area leaders and informed them about the lab. The DCs next facilitated meetings with local government and community leaders in settlements found within these districts; in particular, we focused our recruitment eorts on Kibera slum in Langata District, and Viwandani slum in Embakasi District. The Kibera slum is the largest informal settlement in Nairobi and is situated 5 km from the city center and 2 km from Busara's oce son Ngong Road. Its population was estimated at 170,000 in the 2009 National Census. The slum is divided into a 9 smaller villages:

Kianda, Soweto East, Gatwekera, Kisumu Ndogo, Lindi, Laini Saba, Siranga, Makina, and Mashimoni. Viwandani is a smaller slum is located 7 km from the city center, in Nairobi's industrial area. Established in 1973 on land that was left by the City Council as reserve area on the bank of Ngong River, Viwandani is divided into 5 villages: Paradise, Jamaica, Lunga Lunga, Donholm, and Kingston. Pairs of recruitment ocers moved from village to village in each of the slums over the course of several weeks between April-July 2012. In each area, two community leaders were recruited to mobilize the community. On recruitment days, potential participants rst received a short presentation about the lab, its mission, and their opportunities for study participation. Interested participants read and signed a consent form (illiterate participants were read the consent form and gave their ngerprints to indicate consent), and were signed up for the database by providing their names, contact phone numbers, and some demographic information. In particular, for each participant, we recorded name, age, ethnicity, marital status, number of children, and education level. In addition, a ngerprint was scanned for each respondent, both as a unique identier that obviates the need to require respondents to have national IDs, and to alleviate security concerns when participants are called to the lab to participate in studies. The eligibility requirements consisted in being above the age of 18, and having access to a cell phone and mobile money system MPesa, which was used for payment. Although the penetration of both cell phones and MPesa registration (which is free) is extremely high in the Nairobi slums, respondents who were not signed up for MPesa were allowed to provide the number of a friend or family member. (This occurred in a minority of cases.) Together, these criteria ensure that participation was not limited by socio-economic status, and in particular ownership of a phone, Safaricom line, a national ID, or an MPesa account. In addition to the slum sample, Busara has recently established contacts with local

11

universities to expand recruitment to university students, with the aim of diversifying the subject pool available provided for research. Formal permissions for recruitment at the

University of Nairobi was obtained from the Deputy Vice Chancellor for Student Aairs, and the Deputy Vice Chacellor for Research and Extension Programs, and the Dean of Students provided logistical support. Recruitment of the university sample is ongoing.

Subject pool characteristics

Busara's led ocers set up desks

As is typical for subject pools in experimental and behavioral economics, participants are recruited with a convenience sampling methodology.

in dierent areas of the slums and at the University of Nairobi, in places that are visible and easily accessible, while community mobilizers from the particular slum area brought potential respondents to register. of slums residents. The subject pool thus represents a community sample

For each registered respondent, the database contains information on

gender, age, education, ethnicity, marital status, and number of children. Recruitment of subjects for studies can be stratied according to these criteria. In addition, in addition,

phone number and ngerprints were recorded for communication and sign-up before a study, identication at the gate, and payment after the study. Between April-July 2012, we recruited 1973 potential participants in the Nairobi slums, and 200 potential participants at the University of Nairobi. This paper focuses on the slum sample. 1 illustrates in detail the area of residence of the slum residents registered in Busara's subject pool. Broadly, the large majority of subjects (around two thirds) live in Kibera, and a smaller proportion (one third) were recruited in other slums of Nairobi. Kibera residents are mainly from the Kianda, Makina and Gatwekera villages, while the majority of residents in other slums were recruited in Lunga Lunga, Donholm and Kingston. Table 2 summarizes the gender composition of the Busara subject pool, and compares it to the composition of Nairobi population and Kenyan population. Females are slightly overrepresented in the Busara subject pool (around 54 percent of the sample), possibly because men may be more likely to have jobs that prevent them from participating in studies during working hours. The age of registered subjects ranges from 17 to 93 years, with a mean age of 31.33 (SD: 10.5) for all subjects (N=1973), 30.52 (SD: 10.44) for men (N=910), and 32.03 (SD: 10.59) for women (N=1063). Table 3 summarizes the age composition of the subject pool; the

top panel shows the frequency and percentage of Busara subject pool members in various age groups, while the bottom panel shows the data for Nairobi. Busara's age composition matches that of Nairobi almost exactly; this is even true for older respondents, who might

12

Kibera
Kianda Makina Gatwekera Soweto Raila Mashimoni Lindi Kisumu Ndogo Laini Saba Siranga Undugu Total Kibera Total

Freq.
603 265 215 102 55 30 22 16 9 5 1 1,323

%
30.56 13.43 10.90 5.17 2.79 1.52 1.12 0.81 0.46 0.25 0.05 67.06

Other slums
Lunga Lunga Donholm Kingston Riverside Viwandani Jamaica Dagoretti district Makadara district Kawangware Riruta Other Total other slums

Freq.
260 160 96 65 37 18 4 2 2 2 3 649 1973

%
13.18 8.11 4.87 3.29 1.88 0.91 0.20 0.10 0.10 0.10 0.15 32.89

Table 1: Distribution of the Busara subject pool across slums and villages within slums. Kibera locations are shown in the left columns, other slums on the right.

Busara Gender Freq. %

Male Female Total Table in the the 2: Gender 910 1,063 1,973 46.12 53.88 100 of for

Nairobi Freq. %
1,605,230 1,533,139 3,138,369 the data subject were 51.15 48.85 100

Kenya Freq. %
19,192,458 19,417,639 38,610,097 Busara in the from 49.71 50.29 100 data are and shown and on Kenya

composition data and

pool. Kenya

leftmost

columns,

Nairobi

and

middle

https://opendata.go.ke/Population/Census-Volume-1-Question-1-PopulationHouseholds-a/wd27-eki2
statistics from .

right.

Nairobi

Kenya

obtained

Nairobi

13

Age
18-20 21-30 31-40 41-50 51-60 61-70 71-80 80+ Total 21-30 31-40 41-50 51-60 61-70 71-80 80+ Total

Whole sample Freq. % | > 21

230 912 478 231 97 23 1 1 1,743 940,149 553,761 253,604 105,861 38,188 13,645 8,523 1,913,731 52.32 27.42 13.25 5.57 1.32 0.06 0.06 100 49.13 28.94 13.25 5.53 2.00 0.71 0.45 100

Men Freq. % | > 21 Busara

114 450 211 83 39 12 0 1 796 56.53 26.51 10.43 4.90 1.51 0.00 0.13 100

Women Freq. % | > 21

116 462 267 148 58 11 1 0 947 445,685 317,082 155,010 65,986 22,419 6,886 3,348 1,016,416 48.79 28.19 15.63 6.12 1.16 0.11 0.00 100 43.85 31.20 15.25 6.49 2.21 0.68 0.33 100

Nairobi

494,464 236,679 98,594 39,875 15,769 6,759 5,175 897,315

55.10 26.38 10.99 4.44 1.76 0.75 0.58 100

Table 3: Age composition of the subject pool. columns. Nairobi data were obtained from

Busara data are shown in the top panel,

data for Nairobi in the bottom panel. Separate data for men and women are shown across

Pyramid-Age-Groups-2009/u6e3-zmvi.

https://opendata.go.ke/Population/Nairobi-Pop-

be expected to be less likely to sign up for Busara due to the computerized nature of the studies. However, this appear not to have been a deterrent. Women aged 21-40 are again slightly over-represented in the Busara subject pool, suggesting that, as mentioned above, women of this age group may be more exible with their time than men.

Table 4 shows the ethnic composition of Busara subject pool, and compares it with the population of Kenya as a whole. A large proportion of registered subjects belong to

the Luo, Kikuyu, and Luhya communities, the largest Kenyan tribes. Since a considerable number of Nairobi slums residents are of Luo origin, members of this community tend to be over-represented in Busara subject pool comparing to the Kenyan population. Table 5 displays education attained by registered slum residents, in comparison with Nairobi and Kenya gures. People registered in Busara's subject pool are somewhat more likely to have some level of education, and to have attained secondary education in comparison with the Nairobi and Kenyan population, while also being somewhat less likely to have

14

Busara Ethnicity Freq. %

Luo Kikuyu Luhya Kamba Kisii Meru Kalenjin Taita Embu Maasai Mijikenda Other Total 543 464 359 246 159 22 16 13 8 2 2 138 1,973 27.52 23.52 18.20 12.47 8.06 1.12 0.81 0.66 0.41 0.10 0.10 6.99 100

Kenya Freq. %
1,599 2,088 1,832 952 635 539 1,066 179 228 166 974 1649 11,907 13.43 17.54 15.39 8.00 5.33 4.53 8.95 1.50 1.91 1.39 8.18 7.58 6.27

Table 4: Ethnic composition of the subject pool. Busara data are shown in the left columns, data for Kenya on the right. Kenya data were obtained from the 2008-2009 round of the Kenya Demographic and Household Survey (DHS), a nationally representative dataset on demographic and health indicators for the Kenyan population (http://www.measuredhs.com).

attained primary education only. They are relatively less likely to have university education in comparison to the Nairobi/Kenyan population; this may be due to the fact that tertiary education is dicult to attain for slum residents. Women in our subject pool are relatively more likely than men not to be educated, or to have primary education only; in contrast, they are underrepresented at all other levels of education. This is broadly consistent with the patterns found in the dat for Nairobi and Kenya. An analysis of the marital status of registered slum residents (Table 6) shows that a higher proportion of registered subjects (both women and men) are single in comparison to the Kenyan population; also, the Busara subject pool includes a relatively lower proportion of married/cohabiting subjects than the Kenyan average. Table 7 shows the data for the number of children of registered subjects. The majority of registered women have one to four children, reecting the statistics of the Kenyan population. By contrast, the majority of registered men are relatively more likely not to have children than Kenyan men in general. Together with the data on marital status, this suggests that single men with no children are slightly over-represented in the Busara subject pool. Finally, Table 8 presents the results of a basic arithmetic screen conducted on a randomly selected sample of 38 subject pool members in July 2012. They were asked to give a series of responses on the touchscreen to assess their math skills, which are essential in many paradigms in exprimental and behavioral economics. Each correct response was rewarded

15

Busara Nairobi Kenya Education Level Whole Sample Men Women Men Women Men Women Percentages
none pre-school primary secondary college university Other Total Table 5: in the 100 100 100 2.23 0.41 39.43 48 8.11 1.82 1.1 0.44 27.69 56.59 10.99 3.19 3.2 0.38 49.48 40.64 5.64 0.66 5.53 5.59 37.34 32.61 10.48 7.87 0.28 100 5.57 5.42 36.55 31.97 10.89 9.05 0.28 100 100 100 47.1 31.5 14.7 50.2 22.9 10.7 6.7 16.2

Education levels of the subject pool, as percentages. columns, provide data for Nairobi the and Kenya on the

Busara data are shown separately Nairobi for men were

left

right, level.

and women. used did not obtained from

Empty cells in the Nairobi and Kenya data indicate that the sources data for particular education data

https://opendata.go.ke/Education/Population-3-years-and-above-by-Sex-andHighest-Le/x4e7-whsh;
Kenya data were obtained from the 2008-2009 DHS.

Marital status
Single Married or cohabitating Divorced, separated, widowed Total

Busara Whole sample Women Freq. % Freq. %

881 943 149 1,973 44.65 47.8 7.55 100 418 518 127 1,063 39.32 48.73 11.95 100

Men Freq. %
463 425 22 910 50.88 46.7 2.42 100

Kenya Women Men Freq. % Freq. %

853 4,593 827 6,273 13.6 73.22 13.18 100 598 1,617 121 2,336 25.6 69.22 5.17 100

Table 6: Composition of the subject pool by marital status. Busara data are shown in the left columns, data for Kenya on the right, separately for men and women. Kenya data were obtained from the 2008-2009 DHS.

Busara Whole sample Women Number of children Freq. % Freq. %

none 1 or 2 3 or 4 5 or more Total 592 748 413 220 1,973 30.01 37.91 20.93 11.15 100 160 460 292 151 1,063 15.05 43.27 27.47 14.21 100

Men Freq. %
432 288 121 69 910 47.47 31.65 13.3 7.58 100

Kenya Women Men Freq. % Freq. %

722 2,131 1,858 1,562 6,273 11.51 33.97 29.62 24.9 100 699 688 549 400 2,336 29.92 29.45 23.5 17.12 100

Table 7: Composition of the subject pool by number of children. Busara data are shown in the left columns, data for Kenya on the right, separately for men and women. Kenya data were obtained from the 2008-2009 DHS.

16

Task
1 2 3 4 5 6 7 8 One-digit addition One-digit subtraction One-digit multiplication One-digit division Two-digit addition Two-digit subtraction Two-digit multiplication Two-digit division

Obs. Mean Std. Dev.

38 38 38 38 38 38 38 38 0.84 0.92 0.89 0.76 0.87 0.45 0.13 0.39 0.37 0.27 0.31 0.43 0.34 0.50 0.34 0.50

Table 8: Results of the basic arithmetic screen with a subset of respondents of the subject pool.

with KES 5 (~USD 0.07). Each problem was displayed in writing on the screen; e.g. a onedigit addition problem would be displayed as: 6 + 2 = ?. Below the question, a number pad was displayed on the screen (4x4), containing the numbers 0-9, a red eld labeled Clear, and a green eld labeled OK. The arrangement of the digits on the number display was the same as that on a typical mobile phone; since Kenya has high levels of primary education and all of our participants have mobile phones, they are likely to be intimately familiar with both numbers per se, as well as this particular arrangement. To answer a particular question, participants entered their response by touching the respective digits on the screen, and then conrming with OK. In one-digit addition, two one-digit numbers had to be added, and the result was also a one-digit number; in two-digit addition, two two-digit numbers had to be added, and the result was also a two-digit number. The one-digit and two-digit subtraction, multiplication, and division tasks worked analogously. The results are shown in Table 8. Participants performed better at the one-digit than the two-digit tasks, with an average of 85% correct responses for one-digit and 46% for twodigit tasks. Addition had the highest proportion of correct responses at 86%, followed by subtraction (67%), division (58%), and multiplication (58%). Together, these data suggest that our subject pool members are able to complete simple arithmetic problems. As simple math is a basic ingredient for economic experiments, these data thus provide initial evidence that this population of subjects can participate in such experiments using Busara's setup of touchscreen computers.

17

Special protocols for lab studies at Busara

A central goal of Busara is to enable researchers to conduct computerized behavioral and experimental economics studies with participants from the Nariobi slums. This combination of a high-tech infrastructure with a very special population of respondents has necessitated the development of special methods and protocols, which we summarize briey below.

7.1 Recruitment through SMS and phone calls

At the beginning of a study, a researcher generates a list of potential participants based on the Busara subject pool database, strateed by demographic variables if desired. Once this list has been generated, two recruitment methods ara available to researchers. In the rst, and standard, method, the bulk SMS software FrontlineSMS automatically sends personalized SMS invitations to subsets of participants to invite them to a particular session, indicating date and time of the session, and reminding potential participants of Busara's location. To sign up for participation in the study, participants simply reply YES to the SMS; to decline, they do not reply, or reply NO. Replies are automatically processed and sorted by FrontlineSMS, and a list of conrmed attendees is generated based on the YES-responses. Note that participants' inability to read the text messages received from Busara is rarely an issue because it is common practice for illiterate people to have text messages read out to them by literate friends or family members. The list of conrmed attendees is used on the day of the study when arriving subjects are ngerprinted at the gate to conrm that they are in fact invited for the day's session. The advantages of this recruitment method are that it is a) easy to administer because of its automaticity; convenient for respondents because they usually carry their cell phones with them and can sign up with minimal eort and cost (note, however, that a Yes response is more costly in this recruitment method than in the phone call method described below; sending an SMS costs subjects between KES 1-2, i.e. ~USD 0.02-0.03); and c) cheap, because it can be completed easily and quickly by a single RA, and the cost of sending bulk SMS using Safaricom and FrontlineSMS is ~USD 0.15 for an unlimited number of text messages per day. In the second method, RAs call potential respondents and invite them by phone to participate in sessions. This method is much more labor-intensive and costly than the bulk SMS method, since in our experience it takes a single RAs an entire day to recruit a full session of 20 participants. However, the advantage of this method is that respondents can be pre-screened and selected for baseline criteria that are not contained in the Busara database.

18

SMS Phone
Invited who reply  Yes Invited who attend Attended of those who reply  Yes 55% 51% 70% 48% 33% 67%

Table 9: Percentage of respondents invited for studies at Busara who a) are invited and reply Yes, b) are invited and attend, c) reply Yes and attend, shown separately for subjects recruited through SMS and phone calls. Data are based on

7.2 Fingerprinting for identication

All subject pool members are ngerprinted during recruitment, and the ngerprint information is stored in the subject database. When participants arrive at the gate for a study, they are identied by ngerprint: participants put their thumb on the ngerprint scanner, and the ngerprinting software recognizes them and automatically populates the participant's name and age on the screen of the Identify Netbook. The research assistant welcoming participants at the gate then checks this information against the list of invitees for the particular session to ensure that the participant is invited for that session, and to record their attendance. The attendance information is later added to the subjects database to allow tracking of participation. Fingerprinting has three distinct advantages: rst, in a high-crime city like Nairobi, it serves as a safety precaution; opening up an environment with expensive computer equipment to any participant from the Nairobi slums above the age of 18 potentially creates vulnerabilities to raids and theft, and anecdotal evidence suggests that the fact that Busara keeps ngerprint records of all participants serves as a powerful deterrent against such activities. Second, ngerprinting is superior to all viable alternatives for identication purposes before studies; in particular, although every Kenyan is in principle required to have a National ID, many of our respondents do not own one, and even if they did, requiring them to bring it both for signing up to the subject pool, and to identify themselves at the beginning of each study, would leave to a large number of potential subjects unidentied and unable to participate. Fingerprinting solves this problem completely; in addition, it

is extremely quick, requiring only a few seconds per participant. Finally, ngerprinting at the gate when participants arrive for a study allows us to keep track of which members of our subject pool have participated in which studies; this enables us to track attendance by respondent, study, and signup mechanism (phone vs. SMS), and can later be used to

selectively exclude participants from participation in particular studies depending on their participation in earlier studies.

19

7.3 Touchscreen computers

The central feature of Busara's computer setup that allows us to work with our special subject population is the fact that all lab computers are equipped with touchscreens. During initial testing we put great emphasis on ensuring that participants could actually use these touchscreens. In particular, in July 2012 we test 71 randomly selected subject pool members on a series of simple tasks to be performed on the touchscreen. Specically, participants

were asked to perform the following tasks (cf. Table 10); each correct response was rewarded with KES 5 (~USD 0.07). 1. Tasks 1-4: Two black boxes, each ~3x3 in size, were displayed on a gray screen in a particular conguration. The boxes were either arranged next to each other, or Participants were asked to choose either the left or

stacked on top of each other.

the right box in the former case, or the top or the bottom box in the latter case, by pressing it with their nger. The instruction was given both on the screen in English ("Please put your nger on the {left; right; top; bottom} black area.") and verbally, by a research assistant reading out the same sentence translated to Swahili. Table 10 reports the mean proportion of correct responses and its standard deviation for this task. Participants performed it almost perfectly, with only a few errors. 2. Tasks 5-9: A horizontal row of 5 boxes (each 3x3) was displayed on the screen, each with a dierent color (red, green, black, blue, yellow). In each of 5 consecutive trials, participants were asked to press a certain color. The order of the boxes was randomized with respect to the order of the questions. Again the instruction was given both in English writing on the screen, and orally in Swahili. It can be seen from Table 10 that participants had little trouble with this task. 3. Tasks 10-14: A horizontal row of 5 boxes (each 3x3) was displayed on the screen, each showing a dierent object (cow, hammer, tree, chair, car). In each of 5 consecutive trials, participants were asked to press a certain object. The order of the boxes was randomized with respect to the order of the questions. Again the instruction was given both in English writing on the screen, and orally in Swahili. that participants were able to easily complete this task. Again Table 10 shows

7.4 Subject Instruction

Another important element in running studies at Busara has emerged to be careful instruction of respondents. Our subject pool members are highly unfamiliar with a computer

20

Task
1 2 3 4 5 6 7 8 9 10 11 12 13 14 Choose left Choose right Choose up Choose down Choose red Choose green Choose black Choose blue Choose yellow Choose cow Choose hammer Choose tree Choose chair Choose car

Obs. Mean Std. Dev.

71 71 71 71 71 71 71 71 71 71 71 71 71 71 0.94 0.97 0.99 0.96 0.99 0.96 0.99 1.00 1.00 1.00 0.94 0.99 1.00 0.97 0.23 0.17 0.12 0.20 0.12 0.20 0.12 0.00 0.00 0.00 0.23 0.12 0.00 0.17

Table 10: Results of the touchscreen test performed with a subset of respondents of the subject pool.

lab environment in which they are asked to complete decision-making tasks and participate in economic exchange games, and confusion and misunderstanding is frequently the consequence. Below we summarize the main strategies we employ to facilitate optimal understanding of tasks and games. 1.

Brieng before and after the study.

of participants in the waiting room.

Each lab sessions begins with a brieng

They are welcomed to Busara and told that

they are about to participate in a study about economic behavior and decision-making using touchscreen computers. Invariably participants are worried that they cannot

participate due to their unfamiliarity with computers; this initial brieng procedure has proven useful in alleviating these worries and ensuring participants that they will have few problems in using the comupters, and that help is available for those who require it. At this stage, subjects are also told to switch their phones o, and asked not to talk to each other during the study, but instead raise their hand if they have a question to attract the attention of an RA who will then come to assist them. 2.

Oral explanation by research assistants.

Each experimental session is run by

2-3 RAs; one of them operates the control computer which displays the tasks and paradigms, while the other two instruct the respondents and answer their questions. Specically, a second RA stands by the side of the room and reads through the protocol for the experimental session in Swahili, which contains all instructions for the tasks to be performed, including information which displayed on the screen in English. If

21

respondents have questions, they can raise their hand, and the third second RA comes to their cubicle to assist them. Thus, before respondents make any response, they receive detailed information about the choice they are about to make from three sources: rst, the information displayed on the screen in English; second, the oral explanations given by the RA in Swahili; third, personalized assistance at their workstation if they require it. Together, these measures have proven eective in facilitating respondent

understanding of tasks. 3.

On-screen instruction in writing and through animation and audio.

It has al-

ready been described above that on-screen written instructions in English have proven eective in facilitating understanding when combined with oral explanations by an RA in Swahili. In addition to this technique, we have also used custom-built video For instance, Ed-

animations and audio recordings to explain more comlex tasks.

ward Miguel, Simon Galle, Kelly Zhang and collaborators administered a modied Public Goods Game to participants from our subject pool at Busara in the summer of 2012; beacuse this game was somewhat complex, they employed a custom-made animation video (produced with Microsoft PowerPoint), combined with audio explanations recorded by a Busara RA in Swahili, to explain the game to participants both visually and auditorily. This method of explanation was highly eective in terms of respondent understanding, as well as ecient in terms of the RA time. 4.

Test questions.

After explanations for a task have been given to all respondents,

we frequently probe understanding by presenting questions about the paradigm on the screen which respondents have to answer. Two approaches have been used in this context: in one, participants are required to enter the correct answer to proceed to the actual game; this method is desirable if it is essential that all participants understand the task exactly, but it may require extra explanations by the circulating RA when repsondents cannot advance to the task because they do not know the answer to a probing question. In the other method, participants are free to proceed to the game as soon as they have given a response to the probing question. In this case, it is not guaranteed that they have understood the game entirely, and their data may have to be discounted or excluded later based on this fact; however, this method speeds up study administration considerably because less explaining is required of the RAs. 5.

Practice rounds.

Finally, as is common practice in behavioral economics labs around

the world, nothing claries paradigms to subjects quite as well as a few practice rounds. It is important to demarcate the practice rounds very clearly, both orally and in writing on the screen, and make subject aware when the actual task starts. In our experience,

22

playing 1-5 practice rounds, depending on the game, is highly eective in promoting understanding.

7.5 Special screen displays

Because many of our respondents are unfamiliar with computers, and in particular, mice and keyboards, studies at Busara depend critically on successful use of the touchscreen by subjects. To this end, we have developed several response paradigms which enable subjects to intuitively and easily enter responses on the touchscreen; we describe them briey below. Upon request, zTree les with examples for each of these response formats are available from the authors. 1.

Yes/No questions.

For questions which can be answered with a simple binary

response, subjects are presented the question at the top of the screen in English; in addition, it is read out in Swahili by an RA. Below the question on the screen, a red eld marked No and a green eld marked Yes are shown; respondents are instructed to press the eld whose color corresponds to their answer. Their understanding of colors (see above) suggests that little confusion is likely in making these responses. 2.

Multiple choice questions.

For questions allowing for multiple dierent answers,

subject are presented the question as described above; the answer possibilities appear as a list on the screen, written in English, and are read out by the RA in Swahili. Next to each option is a colored box; the RA instructs subjecst to press the colored box corresponding to the answer of their choice. 3.

Graded responses.

To elicit graded responses from participants  e.g. subjective rat-

ings  we use a 1-100 Likert scale. Again the question is presented as described above; below the written question on the screen, a horizontal bar appears, which shows a color gradient from green to blue. When respondents place their nger anywhere on the gradient, an arrow mark appears at the top of the horizontal bar in the exact location where the respondent placed their nger; in addition, the number corresponding to that particular location (1-100) appears above the horizontal bar. Respondents can change the location of the arrow mark as many times as they wish until the desired location has been selected; then they conrm by pressing a red OK button displayed below the horizontal bar. 4.

Numerical responses.

As described above, numeric responses can be entered by

subjects through a number pad, structured like that of a cell phone, that is displayed

23

on the screen. Subjects enter the number they desire, and then conrm by pressing a green OK button. Corrections can be made by pressing a red Clear button. Due to the widespread use of cell phones in the Nairobi slums, our respondents are intimately familiar with cell-phone like number displays; in addition, basic numeracy is excellent (see above), suggesting that entering numerical information poses few problems to our subjects.

7.6 Payment through MPesa

Finally, after a lab session is completed, subjects receive payment for their participation and, in many cases, their responses in economic games and other incentivized tasks. This payment is transferred to their MPesa accounts immediately after the end of the experimental session through a Safaricom corporate account administered by the Finance Team at IPA-K. To complete the transfer, we rst download the identication data collected on respondents as they enter the session. This data records the computer at which each respondent sat,

and allows us to pay them based on performance and responses, as zTree questions can be assigned payment values and the program will tally up earned payout for each subject individually throughout the session. The identication data also records the phone number the respondent would like payment sent to. For all studies, regardless of whether respondents earned additional money in the games, we pay a base rate of KES 200 (~USD 3) to respondents from Kibera, and KES 400 (~USD 6) to respondents from Viwandani. Each

respondent can earn an additional KES 50 (~USD 0.60) shillings for arriving  on time , i.e. 30 minutes before the intended start of the study. If we overbook a session and respondents who have arrived on time have to be turned away, we pay them the full show up fee and bonus. If a respondent is not on time and less than 90 minutes late, they are reimbursed for the cost of transport at a rate of KES 90 (~USD 1.20) for Kibera respondents and KES 200 (~USD 3) for Viwandani respondents. Respondents who are more than 90 minutes late are paid nothing. It takes on average 2 hours between the end of a session and the receipt of payment in subejcts' MPesa accounts. Once in a respondent's MPesa account, funds can be retrieved at one of many agents found in the Nairobi slums and all around Kenya, or sent directly to other MPesa account holders using MPesa. There are both transfer and withdrawal fees for using MPesa; the fees vary depending on the size of the transfer, and Busara covers only the transfer fees, while withdrawal fees have to be borne by respondents. They are, however, much smaller than the minimum showup fee. MPesa is uniquely suited to the needs of Busara for three reasons. First, virtually all

24

residents of the Nairobi slums use it; recent data on cell phone useage in Kenya by Tavneet Suri suggests that upwards of 90% of Nairobi slum residents have access to MPesa. In

addition, to receive payment for studies completed at Busara, subjects need in fact not have MPesa access themselves; it is sucient if they provide the cell phone number of a friend or relative to whose account the money can be transferred. Thus, using this method of payment excludes virtually nobody from participating in Busara studies. Second, MPesa is much safer than paying in cash: the latter method would require keeping relatively large amounts on-site at Busara, which would additionally increase the risk of raids or theft. Processing payments through MPesa entirely eliminates this risk. In addition, IPA-K's corporate MPesa account is operated with an additional security feature: after payment data is entered into Safaricom's online MPesa interface by the IPA-K nance team in Nairobi immediadely after a session, it rst has to be reviewed and approved by a second member of IPA-K's nance team, working in another location several hundred miles outside of Nairobi. Thus, a raid of the Nairobi

oce alone would not be sucient to force Mpesa transfers to be made. Finally, the use of MPesa for payment is highly convenient from the point of view of particular paradigms in experimental and behavioral economics, in particular those which require delayed payments. For instance, intertemporal choice tasks routinely require payments to be made after a delay; this is not only easy to implement with MPesa, but the fact that both immediate and delayed transfers are made through MPesa additionally controls for a familiar problem in intertemporal choice tasks, namely holding transaction costs constant. With MPesa,

transaction costs are exactly identical, making this payment mechanism superior to many other methods standardly used in this literature.

A Busara Session from Beginning to End

Busara is open to researchers from institutions around the world; we hope that it will provide an infrastructure for economists, psychologists, political scientists, sociologists, anthropologists, and other social scientists to better understand the behavior and preferences of people living in poverty. In the following we summarize a complete experimental session at Busara to give researchers a sense of what they can expect when running studies here. This protocol is exible, however, and can be adapted to diverse experimental needs.

8.1 Overall setup and throughput

A typical study session at Busara Center takes 3 hours on average, and we typically run 2 sessions per day: a morning session (9am-12pm) and an afternoon session (1pm-4pm).

25

Running sessions earlier or later is problematic due to the signicant travel time required in Nairobi for both RAs and respondents; in particular, travel after dark is dangerous, and therefore a timely end to the day's sessions is imperative. Each session can accommodate 20 participants; we typically obtain between 26-28 Yes responses for a given session to ensure that the session gets lled with 20 participants. With 2 sessions a day, this throughput corresponds to a theoretical maximum of 200 participants per week and 800 participants per month. These numbers can decrease somewhat due to unforeseen circumstances that arise during the lifecycle of a study; however, Busara can operate very close to the theoretical capacity maximum: in the summer of 2012, Busara ran 620 subjects between mid-July and mid-August.

8.2 Gate identication procedure

A study session is preceded by participant identication at the gate, which commences 1 hour before the scheduled start time of the study. Respondents are asked to arrive at least 15 minutes before start time; the KES 50 (~USD 0.60) incentive for arriving on time has greatly increased respondent punctuality. Upon arrival at the compound, respondents are identied by ngerprints as described above; specically, the Identify program identies respondents' ngerprints and then reads respondent information from an external database that is pre-loaded before each study with the names and Survey IDs of only those subject pool members who were invited to the day's session. During the identication procedure at the gate, personal information such

as marital status, number of children, phone number, or education level of the invitees is double-checked and updated if necessary. The Identify software also randomly assigns them a place number, and each respondent receives a laminated card with their place number; as described above, these cards are additionally color-coded and marked with symbols, and together the colors and symbols on each card uniquely identify the computer at which the respondent will be sitting. The rationale for randomizing seating is to avoid friends sitting close to each other and potentially colluding during studies. In addition, after identication, each respondent receives a security badge allowing them to enter the compound. They are then escorted by a sta member to the Busara Center waiting room. If more than the maximum number of participants specied for a particular session arrive on time, those subjects who arrive last are turned away; however, they are paid the full showup fee and a bonus. If subjects arrive less than 90 minutes late and have to be turned away, they are paid a transport allowance. If a respondent is late by more than 90 minutes, they receive no payment.

26

8.3 Waiting room protocol

Once all invitees are registered and seated at the waiting room, a welcome speech is given by a Busara sta member. Participants are reminded about the compensation they will receive after the session for their transport and time; they are also told whether the particular session in which they are about to participate will involve tasks in which they can earn additional income. Phones are required to be turned o before participants enter the testing room.

In addition, respondents are not allowed to talk to each other once they have entered the computer lab; if questions arise, they are requested to raise their hands, upon which an RA will come to assist them. We also encourage participants to use the toilets before the session begins. After this welcome brieng, participants are shown into the computer room and

asked to sit down next to the computer with the number indicated on their place cards. Many of our respondents have young children to take care of; we advise these participants to leave their children at home, or bring along someone who will watch over the children during the session. In cases where neither is possible (an average of 1-2 participants per

session of 20), they have the choice of sitting at the workstation with their babies as long as the latter are not disruptive, or of Busara sta watching over the children in the waiting room.

8.4 Computer room protocol

Once in the computer room, consent forms are handed out containing more information about the study; participants have a choice of English or Swahili forms. The Busara RA

running the session goes over the consent form orally in Swahili; after any questions have been answered, participants are asked to sign the form. Participants who are unable to read or write provide ngerprints. After consent forms have been signed and collected, the RA gives an introduction on how to use the touch screen computers to make responses in the experimental tasks. The RA then goes through the study protocol, explaining each task, asking respondents to answer test questions and complete practice rounds. In some cases paradigms are explained through onscreen animations and audio explanations delivered through headphones. Participants who do not understand the tasks or need help using the workstations are assisted by Busara sta. Once the instructions, test questions, and practice rounds are completed, participants perform the actual experimental tasks.

27

8.5 Leaving protocol

Once all experimental tasks are completed, subjects are informed of their nal payout through a display on their screens. The RA thanks them for participation and reminds them that they will be paid through MPesa later the same day. Participants who want to update their mobile payment phone number are asked to do so before they leave. Those who have given third party numbers (relatives, friends, spouses) are reminded to approach the people to whose numbers they asked their payment to be transferred to collect the payment. Participants are then asked to leave their workstations; Busara sta collect the place cards and security tags as the respondents are escorted to the gate. In some studies, a few participants are

randomly selected to remain behind for an exit debrief interview lasting 5-10 minutes.

8.6 Stratication and subject-specic treatments

One of the advantages of recording a number of demographic Recruitment lists are generated in dierent ways depending on the study. We can target a specic pool of people (i.e. those who have completed primary school, are between 20-30 years old, are married, etc.) and

can stratify by any of the demographic variables we collect during recruitment if a study needs a particular composition of respondents. Stratied sampling of course does not lead to a perfectly balanced sample, since to ll a session of 20 respondents we normally invite about 35 people, and the number of people from each stratum who actually attend will vary somewhat. Averaging over a number of sessions, however, we have been able to achieve close to perfect balance for stratied studies. If study treatments vary depending on a particular demographic characteristic, a list matching which treatment they are to receive to the client number of the computer they are participating on must be made at the start of each session. When participants arrive at the session, their thumbprint is scanned and they are assigned to a random computer number. Once all participants have arrived, the identication data can be downloaded, converted into Stata format, merged by ngerprint ID with the recruitment data, and then exported again in zTree format, assigning each respondent to the correct treatment. For instance, if treatment varies by gender, the program can produce a .csv le that can be copied and pasted directly into a zTree le to generate subject-specic treatments that vary by demographic variable.

28

Conclusion

Laboratory research in behavioral and experimental economics in developing countries is still in its infancy, and the eort described here is only a rst step towards enabling social scientists to conduct rigorous laboratory studies with respondents from the world's poorest communities. It is our hope that social scientists from around the world will make intensive use of the infrastructure and opportunities that Busara oers, and that the insights generated by this work will bring us closer towards better understanding the behavior and preferences of people living in developing countries. Ultimately, we hope that this knowledge will translate into interventions that can be used to improve their welfare. Researchers interested in

running studies at Busara are encourage to contact the rst author (joha@mit.edu).

29

References
Arnett, J. J. (2008). The neglected 95%: Why american psychology needs to become less american.

American Psychologist 63

(7), 602614.

Cardenas, J. C. and J. Carpenter (2008). Behavioural development economics: Lessons from eld labs in the developing world. 311338.

The Journal of Development Studies 44

(3),

Fischbacher, U. z-tree - experimenter's manual. IEW - Working Paper iewwp021, Institute for Empirical Research in Economics - University of Zurich. Glimcher, P. W. (2003, March). Decisions, uncertainty, and the brain: neuroeconomics. Glimcher, P. W. (2009). The science of

Neuroeconomics: Decision making and the brain

. Academic Press.

Henrich, J., R. Boyd, S. Bowles, C. Camerer, H. Gintis, R. McElreath, and E. Fehr (2001). In search of homo economicus: Experiments in 15 small-scale societies. Artefactual Field Experiments 0059, The Field Experiments Website. Henrich, J., S. J. Heine, and A. Norenzayan (2010a, July). Most people are not WEIRD.

Nature 466

(7302), 2929.

Henrich, J., S. J. Heine, and A. Norenzayan (2010b, July). Most people are not WEIRD.

Nature 466

(7302), 2929.

Kahneman, D. and A. Tversky (1979). Prospect theory: An analysis of decision under risk.

Econometrica: Journal of the Econometric Society

(3), 549570.

, 263291.

Lei, V. and C. N. Noussair (2002). An experimental test of an optimal growth model.

American Economic Review 92 of Psychology 57

(1), 87115.

Schultz, W. (2006). Behavioral theories and the neurophysiology of reward.

Annual Review Journal of

Smith, V. (1982). Microeconomic systems as an experimental science.

nomic Review 72

(5), 923955.

The American Eco-

Smith, V. L. (1962). An experimental study of competitive market behavior.

Political Economy 70

30